Image: Anthropic
Claude Code users have spent another week arguing over whether Anthropic quietly made Claude Opus 5.5 worse. Fresh complaints were still landing on X at the time of this writing. Independent trackers and at least one large private log dump show little or no sustained drop, and Anthropic has not answered the charge in public.
The loudest new evidence came from Shane Melaugh. He said Gemini 4 Argon reviewed about 1,400 of his Claude Code sessions and found a sharp turn in how Opus 5.5 reasoned on November 3, even though the model name stayed the same. In follow ups he described less reasoning before the first action, more errors, and a nosedive in how much reasoning the model did at high effort. His main post had drawn more than 440,000 views and about 2,600 likes at the time of this writing.
is Opus 5.5 nerfed?
— Shane Melaugh (@shanemelaugh) November 7, 2026
I just had Gemini 4 Argon analyze 1,400 Claude Code sessions and it found that on November 3, something definitely happened.
the model name didn't change, but the way it does reasoning took a sharp turn.
Theo dug into his own cliproxy logs after that thread. He wrote that Anthropic appears to be rolling out a change to the reasoning summarizer on a per account basis, that he saw no sign of degraded capability on his side, and that the company may be trying to fix reasoning traces that had looked poor for weeks. Anthropic's own docs say visible thinking for Opus 5.5 is often a summary rather than the full private chain of thought, which makes a summarizer change hard for users to separate from a real capability cut.
Dug into this more. Anthropic appears to be rolling out a change to the reasoning summarizer on a per-account basis.
— Theo - t3.gg (@theo) November 8, 2026
No sign of degraded capability for me. I think they're just trying to fix the reasoning traces (they've been utter garbage for weeks, complained about this yesterday lol)
Other developers still report the opposite. Posts this week describe the model jumping into work too soon, repeating rejected approaches, ignoring notes, and burning through usage faster when it loops. One widely shared complaint said limits drain faster because the model loops wrong decisions. Those reports are anecdotal and vary by account, which matches Theo's per account theory and leaves outsiders without a single shared timeline.
Harder numbers cut the other way. LiveNerf and BridgeBench's NerfBench, cited in recent coverage, have not shown a sustained regression on their frozen panels. A developer who checked about 67,000 Opus messages in personal Claude Code logs reported failed tool calls at 1.9 percent before the alleged change and 1.8 percent after, with tool calls per message unchanged. That is one person's corpus, not a lab study, but it is one of the few before and after counts in public.
people are saying opus 5.5 got nerfed on nov 3, so i checked my own claude code logs
— argofowl (@argofowl) November 8, 2026
67k opus messages, before vs after:
tool calls per message: 1.06 -> 1.06
failed tool calls: 1.9% -> 1.8%
me getting mad at it: 15.7 -> 14.2 times per 100 messages
not seeing a nerf on my account
Claude Code itself shipped harness changes around the same dates. Version 2.1.317, published on November 3, changed subagents and teammates on a different model from the main conversation so they use that model's saved effort level or its default instead of the session's effort level. Version 2.1.318, published on November 4, removed an earlier fix that kept the model's earlier thinking when resuming a conversation started on 2.1.286 or earlier, so that thinking can now be dropped. Neither release notes a weight change for Opus 5.5.
Anthropic's Opus 5.5 docs add another confounder. The model defaults to medium effort, while earlier Opus models defaulted to high, so a request that omits effort runs one level lower. Raising effort is a documented control, and several users say xhigh restored behavior they liked. That does not prove a silent nerf. It does explain why two people on the same model name can feel different results.
The debate echoes the broader nerfed model complaint covered earlier on this site. It also sits beside reporting that Claude Next could be Opus 5.6. At the time of this writing, searches of Anthropic's news page, status incidents, and the @AnthropicAI and @claudeai accounts turned up no public reply that confirms or denies a November change to Opus 5.5. Until Anthropic publishes a serving or summarizer note, users are left comparing private logs, frozen trackers, and client changelogs that never quite answer the same question.
Comments